Equal Provision, Unequal Benefit: Social Class and the Conversion of Employer-Provided Digital Training into Worker Wellbeing Across 27 European Countries

Abstract

Employers increasingly provide training for working with digital technologies and artificial intelligence (AI), but it is unclear whether the same provision benefits all workers equally. Integrating the job demands–resources (JD-R) model with Sen’s capability approach, this study treats workplace AI exposure as a potential job demand, digital skills and employer-provided digital training as resources, capability-relevant wellbeing as the outcome, and social class as a proxy for the conversion conditions under which resources are used to ensure wellbeing. Structural equation models with ordinal indicators were estimated for 8065 employees in 27 EU countries (Special Eurobarometer 101.4), with country fixed effects, country-clustered and delete-one-country jackknife standard errors, and Random Forest and SHAP analyses as robustness checks. AI exposure was weakly and negatively associated with wellbeing (β = −0.073) and explained almost none of the variance in AI threat appraisal (R2 = 0.002). Employer-provided digital training (β = 0.209) and digital skills (β = 0.189) were the strongest workplace correlates of wellbeing and were robust to every specification. With scalar measurement invariance established, multi-group models showed that the association between employer-provided training and wellbeing was weaker among working-class than among middle-class employees (0.072 versus 0.130; Wald p = 0.001, jackknife p = 0.012), and the model explained 12.0% of working-class wellbeing compared with 31.1% among upper-class employees. The pattern suggests that equal provision of employer digital training may not yield equal benefit, although the class difference is consistent in direction rather than statistically conclusive across all specifications.